streamlit_app.py

import streamlit as st
import cv2
import numpy as np
import pandas as pd
import tensorflow as tf

from src.siamese_model import build_siamese_model
from src.inference import verify_images

# ==============================
# PAGE CONFIG
# ==============================
st.set_page_config(
    page_title="Smart Election Verification System",
    page_icon="🗳️",
    layout="centered"
)

st.title("🗳️ Multimodal Biometric verification System")

# ==============================
# GLOBAL CONSTANTS
# ==============================
IMG_SIZE = 128
THRESHOLD = 0.5

# ==============================
# MODE SELECTION (MUST BE FIRST)
# ==============================
MODE = st.selectbox(
    "Select Verification Mode",
    ["Iris Verification", "Fingerprint Verification"]
)

# ==============================
# MODEL LOADING
# ==============================
model_path = (
    "models/iris_siamese.keras"
    if MODE == "Iris Verification"
    else "models/finger_siamese.keras"
)

@st.cache_resource
def load_model(path):
    model = build_siamese_model()
    model.load_weights(path)
    return model

model = load_model(model_path)

# ==============================
# SIDEBAR – DATASET INSPECTOR
# ==============================
st.sidebar.header("📂 Dataset Inspector")

csv_path = (
    "pairs/iris_pairs.csv"
    if MODE == "Iris Verification"
    else "pairs/finger_pairs.csv"
)

if st.sidebar.checkbox("View Pair Dataset"):
    df = pd.read_csv(csv_path)

    st.subheader("📊 Full Pair Dataset")
    st.write(f"Total Pairs: {len(df)}")
    st.write(
        f"Genuine: {df['label'].sum()} | "
        f"Impostor: {len(df) - df['label'].sum()}"
    )

    # ===== INTERACTIVE TABLE =====
    selection = st.dataframe(
        df,
        use_container_width=True,
        selection_mode="single-row",
        on_select="rerun"
    )

    # ===== ROW CLICK → IMAGE VIEW =====
    if selection and selection["selection"]["rows"]:
        row_idx = selection["selection"]["rows"][0]
        row = df.iloc[row_idx]

        img1 = cv2.imread(row["img1"], cv2.IMREAD_GRAYSCALE)
        img2 = cv2.imread(row["img2"], cv2.IMREAD_GRAYSCALE)

        st.subheader("🖼️ Selected Pair Visualization")

        st.image(
            [img1, img2],
            caption=["Image 1", "Image 2"],
            width=220
        )

        st.info(
            "Label: "
            + ("✅ Same Person" if row["label"] == 1 else "❌ Different Person")
        )

# st.sidebar.header("📂 Dataset Inspector")

# df = None  # SAFE DEFAULT

# if st.sidebar.checkbox("View Pair Dataset"):
#     csv_path = (
#         "pairs/iris_pairs.csv"
#         if MODE == "Iris Verification"
#         else "pairs/finger_pairs.csv"
#     )

#     df = pd.read_csv(csv_path)

#     st.subheader("🔍 Dataset Preview")
#     st.write(f"Total Pairs: {len(df)}")
#     st.write(f"Genuine: {df['label'].sum()} | Impostor: {len(df) - df['label'].sum()}")

#     st.dataframe(df.head(20), use_container_width=True)

# if df is not None and st.sidebar.checkbox("View Sample Image Pairs"):
#     sample = df.sample(1).iloc[0]

#     img1 = cv2.imread(sample["img1"], cv2.IMREAD_GRAYSCALE)
#     img2 = cv2.imread(sample["img2"], cv2.IMREAD_GRAYSCALE)

#     st.image([img1, img2], caption=["Image 1", "Image 2"], width=200)
#     st.write("Label:", "Same Person" if sample["label"] == 1 else "Different Person")

# ==============================
# SIDEBAR – MODEL INFO
# ==============================

def get_model_summary(model):
    summary = []
    model.summary(print_fn=lambda x: summary.append(x))
    return "\n".join(summary)

st.divider()
st.subheader("🧠 Model Information")

show_model_info = st.checkbox("Show Model Information")

if show_model_info:

    # ===== Model Architecture =====
    with st.expander("📐 Architecture Summary", expanded=False):
        st.code(get_model_summary(model), language="text")

    # ===== Model Statistics =====
    with st.expander("📊 Model Statistics", expanded=True):
        col1, col2 = st.columns(2)

        with col1:
            st.metric("Input Shape", "128 × 128 × 1")
            st.metric("Embedding Size", "128")

        with col2:
            st.metric("Total Parameters", f"{model.count_params():,}")
            st.metric(
                "Trainable Parameters",
                f"{sum(tf.size(w).numpy() for w in model.trainable_weights):,}"
            )

if st.sidebar.checkbox("Show Model Stats"):
    st.write("**Input Shape:**", model.input_shape)
    st.write("**Output Shape:**", model.output_shape)
    st.write("**Parameters:**", model.count_params())

if st.checkbox("Show Model Diagram"):
    tf.keras.utils.plot_model(
        model,
        show_shapes=True,
        expand_nested=True,
        to_file="siamese_architecture.png"
    )
    st.image("siamese_architecture.png", caption="Siamese Network Architecture")

# ==============================
# SIDEBAR – SYSTEM STATUS
# ==============================
st.sidebar.header("⚙️ System Status")
st.sidebar.write("Mode:", MODE)
st.sidebar.write("Threshold:", THRESHOLD)
st.sidebar.write("Model Loaded:", "✅ Yes")
st.sidebar.write("Framework:", "TensorFlow + Streamlit")

# ==============================
# IMAGE UPLOAD & VERIFICATION
# ==============================
st.subheader("🔐 Identity Verification")

img1_file = st.file_uploader("Upload Registered Image", type=["jpg", "png"])
img2_file = st.file_uploader("Upload Live Image", type=["jpg", "png"])

def preprocess(img):
    img = cv2.resize(img, (IMG_SIZE, IMG_SIZE))
    img = img / 255.0
    img = np.expand_dims(img, axis=-1)
    img = np.expand_dims(img, axis=0)
    return img

if img1_file and img2_file:
    img1 = cv2.imdecode(
        np.frombuffer(img1_file.read(), np.uint8),
        cv2.IMREAD_GRAYSCALE
    )
    img2 = cv2.imdecode(
        np.frombuffer(img2_file.read(), np.uint8),
        cv2.IMREAD_GRAYSCALE
    )

    p1 = preprocess(img1)
    p2 = preprocess(img2)

    similarity = model.predict([p1, p2])[0][0]
    verified = similarity > THRESHOLD

    st.image([img1, img2], caption=["Registered", "Live"], width=200)
    st.metric("Similarity Score", f"{similarity:.4f}")

    if verified:
        st.success("✅ VERIFIED – VOTE ALLOWED")
    else:
        st.error("❌ NOT VERIFIED – ACCESS DENIED")